Method and device for predicting electric energy use data of user and electronic equipment

Abstract

The invention discloses a user electric energy use data prediction method and device and electronic equipment, the method is applied to the field of data processing, and the method comprises the following steps: collecting target electric energy data of a target user; inputting the target electric energy data into the target prediction model, and calculating target probability distribution of the electric energy use data after the target user responds to the electric power regulation request; and predicting electric energy use data of the target user according to the target probability distribution. Through the method and the device, the problem of relatively low prediction accuracy caused by high sample acquisition cost and limited training sample capacity in the process of predicting the electric energy regulation capability when the user responds to the electric power dispatching request through a neural network model in the related technology is solved.

Description

Technical Field

[0001] The present application relates to the field of data processing, and in particular to a method, device and electronic device for predicting user power usage data. Background Art

[0002] In the actual operation of the power system, the power consumption on the user side is affected by many factors and has great uncertainty. In order to alleviate the supply and demand pressure of the power system, the State Grid Corporation of China has carried out demand response work, and mobilized the adjustment capacity of the user side through economic incentives to achieve the purpose of flexible control of the power generation side and the power consumption side. Therefore, the probabilistic prediction of the demand response adjustment capacity will play an important supporting role in the development of demand response. The uncertainty of the demand response adjustment capacity mainly comes from the user's behavior. At present, researchers mostly use historical data fitting to construct a prediction model for demand response adjustment capacity, but the prediction model of historical data fitting can often only provide a prediction value when predicting, which cannot reflect the uncertainty of the user's adjustment capacity.

[0003] In the process of predicting the user's ability to adjust power when responding to power dispatch requests through neural network models in related technologies, the prediction accuracy is low due to the high cost of sample acquisition and limited training sample capacity. No effective solution has been proposed yet. Summary of the invention

[0004] The main purpose of the present application is to provide a method, device and electronic device for predicting user power usage data, so as to solve the problem of low prediction accuracy due to high sample acquisition cost and limited training sample capacity in the process of predicting the user's ability to adjust power when responding to power dispatching requests through neural network models in related technologies.

[0005] In order to achieve the above-mentioned purpose, according to one aspect of the present application, a method for predicting user power usage data is provided, the method comprising: collecting target power data of a target user, wherein the target power data is power usage data generated after the target user has adjusted power in the past in response to a power regulation request; inputting the target power data into a target prediction model, and calculating a target probability distribution of the power usage data of the target user after responding to the power regulation request, wherein the target prediction model is a model obtained by training a preset prediction model, and the model parameters of the preset prediction model are constructed based on the target power data; and predicting the power usage data of the target user based on the target probability distribution.

[0006] Furthermore, the target prediction model is trained by the following steps: constructing a sample data set based on the historical power data of historical users and the historical behavior data of the historical users; constructing an autoregressive model and a mapping model based on the sample data set, wherein the autoregressive model characterizes the mapping relationship between the historical behavior data of the historical users and the current behavior data, and the mapping model characterizes the mapping relationship between the historical power data and the historical behavior data; inputting the output of the autoregressive model and the output of the mapping model into a preset prediction model to obtain the conditional probability distribution of the power usage data of the historical users after responding to the power dispatch request, wherein the conditional probability distribution includes probability distribution parameters; constructing a loss function of the probability distribution parameters, and using the loss function to optimize the model parameters of the preset prediction model to obtain the target prediction model.

[0007] Furthermore, a sample data set is constructed based on historical power data of historical users and historical behavior data of the historical users, including: acquiring historical power data and historical behavior data, wherein the historical behavior data is behavior data of the historical users in response to power dispatching requests within a preset time period, and the historical power data is power usage data generated after the historical users perform power adjustment in response to power regulation requests; dividing the historical power data into a first data set and a second data set at preset times, wherein the first data set represents known data and the second data set represents data to be predicted; constructing the sample data set based on the historical behavior data, the first data set and the second data set.

[0008] Furthermore, the preset prediction model at least includes: a recurrent neural network, a plurality of feedforward neural networks, and the output of the autoregressive model and the output of the mapping model are input into the preset prediction model to obtain the conditional probability distribution of the power usage data of the historical users after responding to the power dispatch request, including: based on the loop structure in the recurrent neural network, a second hidden layer state is constructed according to the first hidden layer state, the first data set and the historical behavior data, wherein there is a time sequence between the second hidden layer state and the first hidden layer state; the second hidden layer state in the recurrent neural network is mapped to the probability distribution parameters through the multiple feedforward neural networks; and the conditional probability distribution is determined based on the probability distribution parameters.

[0009] Furthermore, an autoregressive model and a mapping model are constructed based on the sample data set, including: constructing the autoregressive model based on a mapping relationship between the first data set and the second data set; constructing the mapping model based on a mapping relationship between the historical behavior data and the second data set.

[0010] Furthermore, a loss function of the probability distribution parameters is constructed, and the loss function is used to optimize the model parameters of the preset prediction model to obtain the target prediction model, including: constructing the loss function based on the probability distribution parameters and the negative log-likelihood loss function; calculating the loss function value according to the historical behavior data and the loss function; iteratively optimizing the model parameters of the preset prediction model according to the gradient optimization algorithm and the loss function value until the loss function value is within the preset numerical range, thereby obtaining the target prediction model.

[0011] Furthermore, predicting the target user's power usage data based on the target probability distribution includes: determining probability distribution parameters corresponding to the target user based on the target probability distribution; determining a confidence level and a significance level of the target user's response to a power dispatch request; and calculating the target user's power usage data based on the probability distribution parameters corresponding to the target user, the confidence level and the significance level.

[0012] Furthermore, the historical behavior data includes at least one of the following: demand response period, incentive price, outdoor temperature, and user preference information.

[0013] In order to achieve the above-mentioned purpose, according to another aspect of the present application, a device for predicting user power usage data is provided, and the device includes: a collection unit, used to collect target power data of a target user, wherein the target power data is the power usage data generated by the target user after performing power adjustment in response to a power regulation request in the past; a first calculation unit, used to input the target power data into a target prediction model, and calculate the target probability distribution of the power usage data of the target user after responding to the power regulation request, wherein the target prediction model is a model obtained by training a preset prediction model, and the model parameters of the preset prediction model are constructed based on the target power data; a prediction unit, used to predict the power usage data of the target user based on the target probability distribution.

[0014] Furthermore, the device also includes: a first construction unit, used to construct a sample data set based on the historical power data of the historical user and the historical behavior data of the historical user; a second construction unit, used to construct an autoregressive model and a mapping model based on the sample data set, wherein the autoregressive model characterizes the mapping relationship between the historical behavior data of the historical user and the current behavior data, and the mapping model characterizes the mapping relationship between the historical power data and the historical behavior data; a second calculation unit, used to input the output of the autoregressive model and the output of the mapping model into a preset prediction model to obtain the conditional probability distribution of the power usage data of the historical user after responding to the power dispatch request, wherein the conditional probability distribution includes probability distribution parameters; a third construction unit, used to construct a loss function of the probability distribution parameters, and use the loss function to optimize the model parameters of the preset prediction model to obtain the target prediction model.

[0015] Furthermore, the first construction unit includes: an acquisition subunit, used to acquire historical electric energy data and historical behavior data, wherein the historical behavior data is the behavior data of the historical users responding to power dispatching requests within a preset time period, and the historical electric energy data is the electric energy usage data generated after the historical users respond to power regulation requests and adjust the power; a division subunit, used to divide the historical electric energy data into a first data set and a second data set according to a preset time, wherein the first data set represents known data and the second data set represents data to be predicted; a first construction subunit, used to construct the sample data set based on the historical behavior data, the first data set and the second data set.

[0016] Furthermore, the preset prediction model at least includes: a recurrent neural network, a plurality of feedforward neural networks, and the second calculation unit includes: a second construction subunit, for constructing a second hidden layer state based on the cyclic structure in the recurrent neural network, according to the first hidden layer state, the first data set and the historical behavior data, wherein there is a time sequence between the second hidden layer state and the first hidden layer state; a mapping subunit, for mapping the second hidden layer state in the recurrent neural network to the probability distribution parameters through the plurality of feedforward neural networks; and a first determination subunit, for determining the conditional probability distribution based on the probability distribution parameters.

[0017] Furthermore, the second construction unit includes: a third construction subunit, used to construct the autoregressive model based on the mapping relationship between the first data set and the second data set; and a fourth construction subunit, used to construct the mapping model based on the mapping relationship between the historical behavior data and the second data set.

[0018] Furthermore, the third construction unit includes: a fifth construction subunit, used to construct the loss function based on the probability distribution parameters and the negative log-likelihood loss function; a first calculation subunit, used to calculate the loss function value based on the historical behavior data and the loss function; an optimization subunit, used to iteratively optimize the model parameters of the preset prediction model based on the gradient optimization algorithm and the loss function value until the loss function value is within a preset numerical range, thereby obtaining the target prediction model.

[0019] Furthermore, the prediction unit includes: a second determination subunit, used to determine the probability distribution parameters corresponding to the target user based on the target probability distribution; a third determination subunit, used to determine the confidence level and significance level of the target user's response to the power dispatch request; and a second calculation subunit, used to calculate the power usage data of the target user based on the probability distribution parameters corresponding to the target user, the confidence level and the significance level.

[0020] Furthermore, the historical behavior data includes at least one of the following: demand response period, incentive price, outdoor temperature, and user preference information.

[0021] In order to achieve the above-mentioned purpose, according to one aspect of the present application, a computer program product is provided, including a computer program, which, when executed by a processor, implements any of the above-mentioned methods for predicting user power usage data, and which, when executed by a processor, implements the steps of the method for predicting user power usage data described in each embodiment of the present application.

[0022] In order to achieve the above-mentioned purpose, according to one aspect of the present application, a computer-readable storage medium is provided, wherein the computer-readable storage medium includes stored computer instructions, wherein when the computer instructions are executed by a processor, any one of the above-mentioned methods for predicting user power usage data is implemented.

[0023] In order to achieve the above-mentioned purpose, according to one aspect of the present application, an electronic device is provided, including one or more processors and a memory, the memory being used to store one or more programs, wherein when the one or more programs are executed by one or more processors, the one or more processors implement any of the above-mentioned methods for predicting user power usage data.

[0024] Through the present application, the following steps are adopted: collecting target power data of the target user, wherein the target power data is the power usage data generated by the target user after the target user has adjusted the power in the past in response to a power regulation request; inputting the target power data into a target prediction model, and calculating the target probability distribution of the power usage data of the target user after responding to the power regulation request, wherein the target prediction model is a model obtained by training a preset prediction model, and the model parameters of the preset prediction model are constructed based on the target power data; predicting the power usage data of the target user based on the target probability distribution solves the problem of low prediction accuracy due to high sample acquisition cost and limited training sample capacity in the process of predicting the user's ability to adjust power when responding to a power scheduling request through a neural network model in the related art. By collecting the target power data of the target user to be predicted in the historical demand response events, it is possible to realize personalized prediction of the response behavior of the target user, and more accurately reflect the power usage data of the target user in response to the power dispatch request. At the same time, by using the target power data to train the preset prediction model, the trained target prediction model is used to calculate the target probability distribution of the power usage data, so that the model can better fit the pattern of changes in the user's power usage, so as to more accurately reflect the user's demand response potential, and can quantify the uncertainty of the change in power usage when the user responds to the power regulation request, and improve the reliability of the prediction, so that the power company can more effectively dispatch power resources and smooth the fluctuations in supply and demand, thereby achieving the effect of improving the operating efficiency and stability of the power system. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] The drawings constituting a part of the present application are used to provide a further understanding of the present application. The illustrative embodiments and descriptions of the present application are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0026] Figure 1 is a flow chart of a method for predicting user power usage data provided in Example 1 of the present application;

[0027] Figure 2 It is a schematic diagram of an optional process for predicting the electric energy usage data of a target user in response to electric power dispatching demand provided in Embodiment 1 of the present application;

[0028] Figure 3 is a schematic diagram of a device for predicting user power usage data provided in Embodiment 2 of the present application;

[0029] Figure 4 It is a schematic diagram of an electronic device for predicting user power usage data provided in Example 5 of the present application. DETAILED DESCRIPTION

[0030] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0031] It should be noted that the user information (including but not limited to user device information, user personal information, collected data, used data, generated data, processed data, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, collected information, used information, generated information, processed information, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, storage, use, processing, transmission, provision, disclosure and application of relevant data are in compliance with the relevant laws, regulations and standards of relevant countries and regions, necessary confidentiality measures are taken, and public order and good morals are not violated, and corresponding operation portals are provided for users to choose to authorize or refuse. For example, an interface is set between this system and relevant users or organizations. Before obtaining relevant information, it is necessary to send an acquisition request to the aforementioned user or organization through the interface, and obtain relevant information after receiving the consent information fed back by the aforementioned user or organization.

[0032] It should be noted that this application provides users with corresponding operation entrances for them to choose to agree or reject the automated decision-making results; if the user chooses to reject, the expert decision-making process will be entered.

[0033] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present application.

[0034] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequential order. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present application described here. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0035] Embodiment 1

[0036] The present invention is described below in conjunction with preferred implementation steps. Figure 1 is a flow chart of a method for predicting user power usage data according to the first embodiment of the present application. Figure 1 As shown, the method comprises the following steps:

[0037] Step S101 : collecting target power data of a target user, wherein the target power data is power usage data generated by the target user after performing power adjustment in response to a power adjustment request in the past.

[0038] At present, when the power system faces an imbalance between supply and demand, the power company will initiate a demand response plan (i.e., the power regulation request mentioned above) to users, inviting them to respond to the demand response plan. Users who respond to the demand response plan will adjust their power usage according to the instructions of the power company, such as reducing power consumption, reducing the use time of high-power electrical appliances, or using renewable energy.

[0039] In the first embodiment of the present invention, it is necessary to collect the target user's past data on adjusting his / her electricity usage according to the power company's power regulation request. The power company or other data collection agencies will record the user's actual electricity usage, and these data constitute the target electricity data. The target electricity data here is a specific numerical record of the change in the user's electricity usage when the user responds to the power regulation request (for example, reducing electricity consumption during peak hours or increasing electricity consumption during off-peak hours). The target user refers to the user who has responded to the power company's power regulation request and is to be predicted to adjust his / her electricity data in response to the power regulation request.

[0040] For example, suppose that the power company initiates a power regulation request to the user, requiring the user to reduce power consumption between 2pm and 4pm, and proposes an incentive condition of rewarding the user with 20 yuan for responding to the power regulation request. The user responds to this power regulation request, turns up the temperature of the air conditioner, and turns off some non-essential lights and household appliances. In this case, the actual reduction in power consumption by the user between 2pm and 4pm (which may be relative to the baseline power consumption before responding to the power regulation request or the average power consumption in a specific time period) is the target power data collected.

[0041] By collecting target power data, the actual response degree of users to power regulation requests can be reflected, including changes in power usage under different incentive conditions, time periods, and external environmental factors. This allows the model to learn the patterns of user response behavior and predict users' power usage in future power demand response events, thereby helping power companies to more accurately estimate the potential for demand response and optimize power dispatch and resource allocation.

[0042] Step S102, input the target power data into the target prediction model, and calculate the target probability distribution of the power usage data of the target user after responding to the power regulation request, wherein the target prediction model is a model obtained by training a preset prediction model, and the model parameters of the preset prediction model are constructed based on the target power data.

[0043] In the first embodiment, the target prediction model is a model for predicting the probability distribution of changes in user power usage, obtained by training a preset prediction model based on target power data and other response behavior-related covariates of the target user. The preset prediction model can be a sequence model such as a recurrent neural network (RNN), a long short-term memory network (LSTM), a gated recurrent unit (GRU), or can be combined with other machine learning algorithms. The goal of training is to adjust the model parameters so that the model can accurately fit the probability distribution of changes in power usage of the target user after responding to a power regulation request.

[0044] The target probability distribution output by the target prediction model provides multiple possible values ​​of the change in power usage and their corresponding probabilities. For example, if the target prediction model predicts that the change in power usage follows a normal distribution, then this distribution will include a mean (representing the most likely change in power usage) and a variance (representing the uncertainty of the change). This not only gives a single predicted value for power usage, but also provides the range of change and the level of uncertainty of the change.

[0045] By integrating the target users' target electricity usage data and user response behavior into a prediction model, the model is trained to learn the user's demand response pattern, and the model is used to predict the user's electricity usage change distribution under future power regulation requests, providing data support and decision-making basis for power demand management.

[0046] Step S103: predicting the target user's power usage data according to the target probability distribution.

[0047] In the first embodiment, a prediction of power usage data can be generated based on the target probability distribution. Specifically, the predicted value of power usage can be obtained from the target probability distribution by sampling (such as Monte Carlo sampling), or the prediction of power usage can be directly obtained by calculating the expected value (i.e., mean) of the probability distribution. In addition, different confidence levels can be set to calculate the confidence interval of the change in power usage at the confidence level, which helps the power company understand the possible range of user power usage at a specific confidence level.

[0048] In power system dispatch, predicting the target user's energy usage data (especially the change in energy usage after responding to power regulation requests) plays a key role in optimizing the balance of power supply and demand, improving system stability, and promoting the consumption of renewable energy. Through prediction based on target probability distribution, power companies can more accurately estimate the potential for demand response, reasonably arrange power resources, reduce the operating risks of the power system, and improve the efficiency and flexibility of dispatch.

[0049] In summary, the method for predicting user power usage data provided in Example 1 of the present application collects target power data of the target user, wherein the target power data is the power usage data generated after the target user has adjusted power in the past in response to a power regulation request; the target power data is input into a target prediction model, and the target probability distribution of the power usage data of the target user after responding to the power regulation request is calculated, wherein the target prediction model is a model obtained by training a preset prediction model, and the model parameters of the preset prediction model are constructed based on the target power data; the power usage data of the target user is predicted based on the target probability distribution, thereby solving the problem of low prediction accuracy due to high sample acquisition cost and limited training sample capacity in the process of predicting the user's ability to adjust power when responding to a power scheduling request through a neural network model in the related art. By collecting the target power data of the target user to be predicted in the historical demand response events, it is possible to realize personalized prediction of the response behavior of the target user, and more accurately reflect the power usage data of the target user in response to the power dispatch request. At the same time, by using the target power data to train the preset prediction model, the trained target prediction model is used to calculate the target probability distribution of the power usage data, so that the model can better fit the pattern of changes in the user's power usage, so as to more accurately reflect the user's demand response potential, and can quantify the uncertainty of the change in power usage when the user responds to the power regulation request, and improve the reliability of the prediction, so that the power company can dispatch power resources more effectively, thereby achieving the effect of improving the operating efficiency and stability of the power system.

[0050] Optionally, in the method for predicting user electricity usage data provided in Example 1 of the present application, the above-mentioned target prediction model is trained by the following steps: constructing a sample data set based on the historical electricity data of historical users and the historical behavior data of historical users; constructing an autoregressive model and a mapping model based on the sample data set, wherein the autoregressive model characterizes the mapping relationship between the historical behavior data of historical users and the current behavior data, and the mapping model characterizes the mapping relationship between historical electricity data and historical behavior data; inputting the output of the autoregressive model and the output of the mapping model into a preset prediction model to obtain the conditional probability distribution of the electricity usage data of historical users after responding to power dispatch requests, wherein the conditional probability distribution includes probability distribution parameters; constructing a loss function of the probability distribution parameters, and using the loss function to optimize the model parameters of the preset prediction model to obtain the target prediction model.

[0051] In an optional embodiment, in order to train the target prediction model, it is first necessary to collect and organize the historical power data and historical behavior data of historical users. Historical power data refers to the records of power usage of multiple users in the past, while historical behavior data includes the responses of multiple users when receiving power dispatch requests, such as demand response period, incentive price, outdoor temperature, user preferences and other covariate information that affects user power usage decisions. These historical power data and historical behavior data are integrated into a sample data set as input for subsequent target prediction model training.

[0052] Then, based on the sample data set, two models are constructed to capture different relationships in historical data. The autoregressive model aims to learn the mapping relationship between the user's historical behavior data and current behavior data, that is, to predict the user's current or future response behavior based on the user's past response behavior. The mapping model is to establish a connection between historical power data and historical behavior data, and predict the specific behavior changes of users in response to power dispatching needs through the user's past historical power data, such as the increase or decrease in the user's power usage.

[0053] Secondly, the output results of the autoregressive model and the mapping model are input into a preset prediction model, where a prediction model is preferably a recurrent neural network (RNN), and the prediction model can also be a long short-term memory network (LSTM) or other suitable models. The purpose of this model fusion is to obtain the conditional probability distribution, that is, the probability distribution of the power usage data after the user responds to the power dispatch request given the user's historical power data and historical behavior data. The conditional probability distribution contains multiple possible values ​​of the predicted power usage data change and their corresponding probabilities, and this distribution is described by model parameters, such as mean, variance, etc.

[0054] Finally, in order to make the preset prediction model more accurately fit the changes in power usage data after historical users respond to power dispatch requests, a loss function needs to be constructed. The loss function measures the difference between the probability distribution parameters predicted by the model and the actual power usage data, and optimizes the model parameters by minimizing this difference. During the model training process, the gradient descent method or other optimization algorithms are usually used to update the model parameters to minimize the loss function value, thereby obtaining the above-mentioned target prediction model.

[0055] By integrating the user's historical power data and historical behavior data, and adopting the fusion of the autoregressive model and the mapping model, the target prediction model can learn the complex patterns of user demand response behavior and accurately predict the changes in power usage that users may make under future power dispatch requests, including the uncertainty description of the changes. This not only improves the accuracy of the prediction, but also takes into account the uncertainty of the prediction, providing power companies with a comprehensive and reliable demand response capability prediction tool, which helps to optimize power dispatch strategies and improve the operating efficiency and stability of the power system.

[0056] Optionally, in the method for predicting user power usage data provided in Example 1 of the present application, a sample data set is constructed based on historical power data of historical users and historical behavior data of historical users, including: acquiring historical power data and historical behavior data, wherein the historical behavior data is behavior data of historical users responding to power dispatching requests within a preset time period, and the historical power data is power usage data generated after historical users responding to power regulation requests and performing power adjustment; dividing the historical power data into a first data set and a second data set at preset times, wherein the first data set represents known data and the second data set represents data to be predicted; and constructing a sample data set based on the historical behavior data, the first data set, and the second data set.

[0057] In an optional embodiment, first, the power usage data of historical users in a preset time period (for example, 1 year, 3 years, 5 years) are extracted. These power usage data record the actual power consumption of historical users at different time points. At the same time, the behavioral data of historical users in response to power dispatch requests during this period are collected, including but not limited to: demand response period, incentive price, external environmental conditions (such as weather) and user preferences. These behavioral data reflect the response behavior of users under different conditions and are key information for the prediction model to understand the user demand response pattern.

[0058] For example, for a user A who has performed N demand responses, obtain the demand response adjustment capability sample D for each time. n , n = 1, ..., N, forming a regulation capability sample set, that is, the above historical power data, recorded as Among them, Dn Represents the power usage data of the user in response to the nth power dispatch demand. Obtain the covariate data X related to the user's response behavior each time the user performs demand response n , forming a covariate data set, that is, the above historical behavior data, recorded as X n ∈R M Among them, X n It represents the covariates related to the user's nth response to the power dispatch demand, including the demand response period, incentive price, outdoor temperature, user preferences obtained by the questionnaire, etc. M is the covariate dimension.

[0059] Then, the acquired historical power data is divided into a first data set and a second data set according to the characteristics of the time series. For example, for a user A who has performed N demand responses, the first data set contains the power usage data of user A from the 1st to the n0-1th time of responding to power dispatch, recorded as n0 is an integer greater than 1 and less than N. The first data set is regarded as known data and is used to build the model and train the model parameters. The second data set contains the power usage data of user A from the n0th to the Nth time of responding to power dispatch, denoted as The second data set is used for validation and testing of the model to evaluate the model's ability to predict future data, ie, data to be predicted.

[0060] Finally, a sample data set is constructed based on the historical behavior data, the first data set, and the second data set. The construction of the sample data set is crucial for model training. It associates the historical behavior data with the corresponding power usage data to form a training sample. For example, for each historical demand response event, the sample data set may include the time of the event, the incentive price, the external environmental conditions, the user's preferences (historical behavior data), and the change in the user's power usage after the response event (historical power data). These sample data sets will be used for subsequent model training to help the model learn the mapping relationship between user demand response behavior and changes in power usage.

[0061] Through the above steps, the constructed sample data set can fully reflect the historical user's electricity usage behavior and its changing rules under power regulation requests. This division method based on time series data helps to improve the generalization ability and prediction accuracy of the prediction model. The constructed sample data set provides sufficient information for model training, enabling the model to learn users' electricity usage patterns and demand response behaviors based on historical data, so that when faced with new power regulation requests, it can accurately predict changes in users' electricity usage and support the optimal scheduling and resource management of the power system.

[0062] Optionally, in the method for predicting user electricity usage data provided in Example 1 of the present application, an autoregressive model and a mapping model are constructed based on a sample data set, including: constructing an autoregressive model based on a mapping relationship between a first data set and a second data set; and constructing a mapping model based on a mapping relationship between historical behavior data and the second data set.

[0063] The autoregressive model is a statistical prediction method that uses historical power usage data (i.e., the first data set) to predict future power usage data (i.e., the second data set). The autoregressive model constructs a prediction model by analyzing the dynamic relationship between the first data set and the second data set, i.e., the autocorrelation characteristics of the power usage data over time. The autoregressive model attempts to capture and learn the inherent laws of the evolution of power usage data over time, so that future power usage can be predicted given historical data. This method is particularly suitable for processing data with time series properties, such as power load data.

[0064] The mapping model establishes the relationship between historical behavior data and a second data set (i.e., the power usage data to be predicted). The historical behavior data contains various information about the user when responding to power regulation requests, such as incentive prices, demand response time periods, external environmental conditions, and user preferences. The purpose of the mapping model is to learn the mapping relationship between these behavior data and changes in user power usage, that is, how to predict the user's future power usage based on current incentive conditions and user behavior. This model can transform covariate information into a prediction of power usage data, thereby taking into account the diversity of user behavior and the impact of the external environment when making predictions.

[0065] For example, when predicting the n0th demand response adjustment capability of the individual user, the known historical demand response adjustment capability (i.e., the first data set ) and the n0th demand response regulation capability The autoregressive model is constructed between . The covariate corresponding to the n0th demand response adjustment capacity is used The ability to respond to n0th demand The mapping model is constructed between them, and the two models are integrated to obtain the conditional probability distribution of the n0th demand response regulation capability, which is recorded as At this point, for all the demand response potentials to be predicted There is a relationship as shown in Formula 1,

[0066]

[0067] in, represents the conditional probability of the power usage data of the user responding to the power dispatch request for the nth time based on the known data (i.e., the first data set) and the historical behavior data, n∈(n0,N), Represents the conditional probability distribution of the power usage data of the user responding to the power dispatch request for the n0th time.

[0068] By combining the autoregressive model and the mapping model, the autoregressive model learns the time series characteristics of the electricity usage data, and the mapping model considers the impact of user behavior and external conditions on electricity usage. This fusion method enables the prediction model to capture the historical patterns of user electricity usage and take into account the immediate changes in user behavior and environmental conditions, thereby improving the comprehensiveness and accuracy of the prediction. The constructed autoregressive model and mapping model analyze the user's electricity usage data from the perspectives of time series characteristics and covariate relationships, respectively. Finally, by integrating the outputs of these two types of models, the probabilistic prediction results of the user's electricity usage data after receiving the power dispatch request are obtained, providing the power dispatch department with more accurate and comprehensive demand response potential prediction information. These prediction results can be used to guide the operation optimization of the power system, including load management, resource allocation, and the formulation of power supply and demand balance strategies, thereby improving the operation efficiency of the power system and its ability to cope with complex market conditions.

[0069] Optionally, in the method for predicting user electricity usage data provided in Example 1 of the present application, the above-mentioned preset prediction model includes at least: a recurrent neural network, multiple feedforward neural networks, and the output of the autoregressive model and the output of the mapping model are input into the preset prediction model to obtain the conditional probability distribution of historical user electricity usage data after responding to power dispatch requests, including: based on the cyclic structure in the recurrent neural network, a second hidden layer state is constructed according to the first hidden layer state, the first data set and the historical behavior data, wherein there is a time sequence between the second hidden layer state and the first hidden layer state; the second hidden layer state in the recurrent neural network is mapped to probability distribution parameters through multiple feedforward neural networks; and the conditional probability distribution is determined based on the probability distribution parameters.

[0070] In the prior art, a recurrent neural network (RNN) is a neural network that can process sequence data. It is very effective in processing time series data because it can remember previously input information and combine this information with current input, which is especially important for power usage prediction because power usage usually exhibits time series characteristics.

[0071] In the first embodiment of the present invention, the loop structure of RNN is used to construct a hidden layer state containing time series information, so as to reflect the dynamic changes of historical user power usage habits and demand response behaviors. Through the loop unit in RNN, the first hidden layer state (i.e., the hidden state at the previous moment, or the power usage data of the n0-1th user demand response), the first data set (i.e., the known data in the historical power data) and the historical behavior data (including the demand response period, the incentive price, the outdoor temperature, the user's preference, etc.) are used as input to construct the second hidden layer state (i.e., the hidden state at the current moment, or the power usage data of the n0th user demand response). There is a time sequence between the first hidden layer state and the second hidden layer state, that is, the latter depends on the former, which reflects the changes and influences of the user's power usage behavior over time.

[0072] For example, the conditional part of the above conditional probability is transformed into Transformed into the hidden layer state of RNN, as shown in Formula 2 and Formula 3,

[0073]

[0074] Among them, h n Represents the current hidden layer state of the RNN network, which depends on the power usage data D of the n-1th user demand response n-1 , the covariate X corresponding to the nth demand response regulation capability n , and the previous hidden layer state h n-1 . The network function representing the RNN recurrent structure (or unit), θ RNN Represents the model parameters of the RNN unit.

[0075] Then, after the hidden layer states of the recurrent neural network are established, multiple feedforward neural networks are used to map these states into probability distribution parameters, such as the mean and variance of the normal distribution. The feedforward neural network can directly map the input data to the output, and is used to extract the statistical features describing the changes in power usage, that is, the parameters of the probability distribution, from the hidden layer states of the recurrent neural network.

[0076] For example, two feedforward neural networks are constructed, and the hidden layer state h n Mapped to the probability distribution parameter μ of the nth demand response regulation capability n and σ n , as shown in Formula 4 and Formula 5,

[0077]

[0078] in, and Represent the network functions of two feedforward neural networks, θ FC1 and θ FC2 Represents the parameters of two feedforward neural networks.

[0079] Finally, the obtained probability distribution parameters are used to determine the conditional probability distribution of the power usage data of historical users after the power dispatch request. The conditional probability distribution refers to the possible values ​​of the power usage data and the probability of its occurrence under certain given conditions (for example, the user's historical behavior data and historical power data). This distribution incorporates the uncertainty of the user's power usage changes into the prediction model, so that the prediction results not only include the most likely power usage change, but also provide the possible range and probability of this change, so that a confidence interval can be given.

[0080] By integrating the memory capacity of recurrent neural networks and the mapping capacity of multiple feedforward neural networks, the preset prediction model can capture the time series characteristics and conditional probability distribution information of users' electricity usage behavior, and achieve accurate prediction of users' demand response potential, including its uncertainty description. This method not only improves the accuracy and reliability of the prediction, but also provides the power dispatching department with confidence interval predictions of changes in user electricity usage, helping them make better dispatching decisions to cope with fluctuations and uncertainties in electricity supply and demand, and optimize the operating efficiency and stability of the power system.

[0081] Optionally, in the method for predicting user electricity usage data provided in Example 1 of the present application, a loss function of probability distribution parameters is constructed, and the loss function is used to optimize the model parameters of a preset prediction model to obtain a target prediction model, including: constructing a loss function based on probability distribution parameters and a negative log-likelihood loss function; calculating the loss function value based on historical behavior data and the loss function; iteratively optimizing the model parameters of the preset prediction model based on a gradient optimization algorithm and the loss function value until the loss function value is within a preset numerical range to obtain a target prediction model.

[0082] In the first embodiment, the negative log likelihood (NLL) loss function is used to construct the loss function. The negative log likelihood loss function is suitable for probability distribution prediction, and its core is to evaluate the degree of match between the probability distribution generated by the model and the actual observed data. By comparing the probability distribution parameters predicted by the model with the actual power usage data in the historical data, the loss function can quantify the deviation between the uncertainty of the prediction and the actual result. Exemplarily, the loss function can be shown as Formula 6,

[0083]

[0084] Among them, L represents the loss function, θ RNN represents the model parameters of the RNN unit, θFC1 and θ FC2 Denotes the parameters of two feedforward neural networks, D n Represents historical electric energy data, μ n ,σ n represents the probability distribution parameters.

[0085] Then, after constructing the loss function, it is necessary to calculate the loss function value based on the historical behavior data and historical power data, that is, substitute the historical user power usage data into the probability distribution predicted by the model, calculate the log-likelihood value of the actual data under the predicted probability distribution, and then take its negative value to obtain a value representing the degree of mismatch between the model prediction and the actual data. The size of the loss function value reflects the accuracy of the model prediction. The smaller the value, the higher the match between the model prediction and the actual data.

[0086] Secondly, a gradient optimization algorithm (e.g., gradient descent, stochastic gradient descent, etc.) is used to iteratively update the model parameters of the preset prediction model according to the calculated loss function value. The goal of the gradient optimization algorithm is to find the model parameters that minimize the loss function value. This process usually calculates the gradient of the loss function with respect to the model parameters, and then adjusts the parameters in the opposite direction of the gradient to gradually reduce the value of the loss function until it converges to a local minimum or reaches a preset numerical range. The process of parameter optimization is the core of model training. Through this process, the model can learn the optimal parameter settings extracted from historical data, so that the model can more accurately predict the user's electricity usage under specific conditions.

[0087] Finally, when the optimization of model parameters reaches the preset convergence standard, that is, when the loss function value stabilizes within a preset numerical range, the training process ends, and the target prediction model obtained is the best model after historical data training and parameter optimization. The model parameters of the best model can be expressed as in, represents the optimal parameters of the RNN network, represents the optimal parameters of the two feedforward neural networks, and L represents the loss function value. This model can predict the probability distribution of the user's future response power usage data after receiving a power dispatch request based on the user's historical power usage data and behavior data, including the confidence interval of the prediction results, providing key information for power dispatch and demand-side management.

[0088] By constructing and optimizing the loss function, the model parameters are effectively adjusted, thereby improving the prediction accuracy of the prediction model's ability to respond to user demand. This not only helps reduce uncertainty and risk in the power market, but also provides the power dispatching department with more accurate user power usage forecasts, supporting more efficient and robust power resource allocation and dispatching decisions.

[0089] Optionally, in the method for predicting user power usage data provided in Example 1 of the present application, the power usage data of the target user is predicted based on the target probability distribution, including: determining the probability distribution parameters corresponding to the target user based on the target probability distribution; determining the confidence level and significance level of the target user's response to the power dispatch request; and calculating the power usage data of the target user based on the probability distribution parameters, confidence level and significance level corresponding to the target user.

[0090] In the first embodiment, the trained target prediction model is used to input the target power data of the target user into the target prediction model, and the target prediction model outputs the probability distribution parameters of the target user's power usage data, such as the mean and variance of the normal distribution. These parameters describe the possible distribution of the user's future power usage, rather than just a single predicted value, which provides a mathematical description for the uncertainty of the prediction result.

[0091] Then, determine the confidence level and significance level. The confidence level refers to the probability that the prediction result can cover the true value, usually expressed as a percentage. For example, a 95% confidence level means that the predicted confidence interval has a 95% probability of containing the true value. The significance level and the confidence level are complementary to each other and are the reciprocals of the confidence level. For example, if the confidence level is 95%, then the significance level is 5%. The purpose of determining the confidence level and significance level is to give a probability confidence interval when predicting to reflect the uncertainty of the prediction. Exemplarily, the significance level is set to α, α≤1, and the corresponding confidence level is (1-α)%.

[0092] Secondly, according to the probability distribution parameters, confidence level and significance level of the target user's target power data, the confidence interval of the target user's power usage data in response to the power dispatch demand is calculated. Exemplarily, the standard normal distribution table (or Z score table) is used to find the Z score value corresponding to the confidence level, and then the upper and lower limits of the confidence interval are calculated in combination with the parameters of the probability distribution (such as the mean and standard deviation). The confidence interval gives the possible range of the user's power usage and the probability that this range covers the true value.

[0093] For example, when predicting the target user's energy usage data after responding to a power dispatch request, the confidence interval can tell the dispatch department that the user's energy usage may fluctuate within a certain range, and the range and probability of this fluctuation are determined by the model's predictive ability and the uncertainty of the data. This enables the dispatch department to more comprehensively consider the complexity and uncertainty of user behavior when planning grid loads, arranging power generation resources, and formulating demand response strategies, thereby avoiding the potential risks of over-reliance on a single predicted value. This approach improves the stability and efficiency of power system operation and helps to better cope with the volatility of renewable energy and the uncertainty of power demand.

[0094] For example, in the z-score table, the corresponding search The corresponding z-score value is recorded as At this time, the prediction interval of the power usage data of individual users in response to power dispatch requests under the confidence level of (1-α)% is

[0095] Through the above steps, not only the estimated value of the target user's power usage after responding to the power dispatch request is provided, but more importantly, it also provides a description of the uncertainty of this estimate, that is, the confidence interval. The prediction of the confidence interval not only increases the practicality of the prediction, but also provides decision support for the power dispatch department, helping them to consider the uncertainty of the prediction results when making dispatch decisions, so as to develop a more flexible and adaptive dispatch strategy.

[0096] Optionally, in the method for predicting user power usage data provided in Embodiment 1 of the present application, the above-mentioned historical behavior data includes at least one of the following: demand response period, incentive price, outdoor temperature, and user preference information.

[0097] In the first embodiment, historical behavior data refers to a series of related behavior data performed by historical users in response to power dispatching requirements, which is crucial for understanding how users respond to power dispatching requests and predicting their future power usage patterns. Specifically, historical behavior data includes, but is not limited to, demand response period, incentive price, outdoor temperature, user preference information, and other data.

[0098] Demand response time refers to the specific time when users responded to power dispatch demands in the past, such as a day of the week, a period of time within a day, a season, or a special event (e.g., holidays). Demand response time is very important for forecasting because users' power usage patterns and response behaviors are often time-dependent, and different time periods may trigger different power demands or responses to incentives.

[0099] Incentive price refers to the price signal or economic incentive provided by power companies to encourage users to change their electricity usage behavior when users respond to power dispatch demand activities. The level of incentive price directly affects whether users participate in and how they respond to demand response activities, and is a key covariate in the prediction model.

[0100] Outdoor temperature refers to the outdoor temperature at which users responded to power dispatch demands in the past. Weather conditions, especially outdoor temperature, have a significant impact on users' power demand. For example, high or low temperature weather may increase the use of air conditioning or heating equipment, thereby affecting power consumption. Therefore, outdoor temperature is one of the external environmental conditions that is taken into account for users' power usage behavior.

[0101] User preference information includes, but is not limited to: user preferences for electricity usage habits, sensitivity to electricity prices, willingness to participate in demand response activities, etc. User preference information reflects the characteristics and behavioral tendencies of individual users and is an indispensable factor in predicting their demand response capabilities.

[0102] Combining these historical behavior data with historical electricity usage data can build a more comprehensive and detailed user behavior model. By analyzing the user's past demand response behavior, the preset model can learn the changing patterns of the user's electricity usage pattern under specific conditions (such as specific time periods and specific incentive prices), and then predict the user's future response behavior and possible electricity usage when receiving power dispatch requests. This analysis method based on historical data helps to improve the accuracy and practicality of predictions and provide solid data support for power dispatch and demand-side management.

[0103] Optionally, in the first embodiment, the scheme is a schematic diagram of the process of predicting the target user's power usage data in response to the power dispatch demand. Figure 2As shown in Figure 1, historical power data of users participating in demand response activities in the past are collected. These data contain the user's power usage records under different conditions (such as different incentive prices, weather conditions, time periods, etc.). The historical power data is divided into power usage data of users responding to power dispatch requests from the 1st to the n0-1st time and power usage data of users responding to power dispatch from the n0th to the Nth time. Based on the historical power data and historical behavior data, an autoregressive model and a mapping model are constructed respectively. The autoregressive model is mainly used to capture the time series characteristics of power usage data, while the mapping model is used to study the relationship between behavior data (such as incentive prices, outdoor temperature, etc.) and power usage data. The introduction of recurrent neural network (RNN) realizes the fusion of autoregressive model and mapping model, so that the model can simultaneously consider the time series dependence of power usage and the influence of behavior data, thereby improving the comprehensiveness and accuracy of prediction. Two feedforward neural networks are constructed to convert the hidden layer states of RNN into probability distribution parameters of power usage data, namely mean and variance. The feedforward neural network maps the hidden state to probability distribution parameters through multiple layers of nonlinear transformation, providing a mathematical basis for subsequent probabilistic prediction. The loss function of the preset model (i.e., the recurrent neural network and two feedforward neural networks) is constructed according to the negative log-likelihood loss function, and the parameters of the preset model are iteratively updated using optimization algorithms such as the gradient descent method to determine the optimal parameter settings so that the prediction results match the actual historical data to the highest degree, thereby obtaining a target prediction model that can perform probabilistic predictions. When the target prediction model is trained, the power usage data of the target user to be predicted can be input, and the trained target prediction model can be used to predict the probability distribution parameters of the power usage data of the target user, that is, the possible mean and variance of the power usage under the current conditions. Finally, with the probability distribution parameters, statistical methods (such as confidence interval calculation) can be used to determine the probability range of the power usage data of the target user when receiving the current power dispatch request. This step provides the power dispatch department with a power usage forecast at a confidence level, which helps them to more accurately evaluate the user's responsiveness and make reasonable dispatch decisions.

[0104] entire Figure 2 The process clearly explains the whole process from data collection to model construction, training, and finally confidence interval prediction, demonstrating the systematic and scientific nature of the prediction method. This method can help power dispatching departments improve the stability and efficiency of power grid operation through more accurate probabilistic predictions when facing complex and changing power market conditions.

[0105] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0106] Embodiment 2

[0107] The second embodiment of the present application also provides a prediction device for user power usage data. It should be noted that the prediction device for user power usage data of the second embodiment of the present application can be used to execute the prediction method for user power usage data provided in the first embodiment of the present application. The prediction device for user power usage data provided in the second embodiment of the present application is introduced below.

[0108] Figure 3 Schematic diagram of a device for predicting user power usage data according to the second embodiment of the present application. Figure 3 As shown, the device includes: a collection unit 301, a first calculation unit 302 and a prediction unit 303.

[0109] Specifically, the collection unit 301 is used to collect target power data of the target user, wherein the target power data is power usage data generated after the target user has adjusted power in the past in response to a power regulation request.

[0110] The first calculation unit 302 is used to input the target power data into the target prediction model to calculate the target probability distribution of the power usage data of the target user after responding to the power regulation request, wherein the target prediction model is a model obtained by training a preset prediction model, and the model parameters of the preset prediction model are constructed based on the target power data.

[0111] The prediction unit 303 is used to predict the power usage data of the target user according to the target probability distribution.

[0112] The device for predicting user power usage data provided in the second embodiment of the present application collects target power data of the target user through the collection unit 301, wherein the target power data is the power usage data generated after the target user has adjusted power in the past in response to a power regulation request; the first calculation unit 302 inputs the target power data into the target prediction model, and calculates the target probability distribution of the power usage data of the target user after responding to the power regulation request, wherein the target prediction model is a model obtained by training a preset prediction model, and the model parameters of the preset prediction model are constructed based on the target power data; the prediction unit 303 predicts the power usage data of the target user based on the target probability distribution, which solves the problem of low prediction accuracy due to high sample acquisition cost and limited training sample capacity in the process of predicting the user's ability to adjust power when responding to a power dispatch request through a neural network model in the related art. By collecting the target power data of the target user to be predicted in the historical demand response events, it is possible to realize personalized prediction of the response behavior of the target user, and more accurately reflect the power usage data of the target user in response to the power dispatch request. At the same time, by using the target power data to train the preset prediction model, the trained target prediction model is used to calculate the target probability distribution of the power usage data, so that the model can better fit the pattern of changes in the user's power usage, so as to more accurately reflect the user's demand response potential, and can quantify the uncertainty of the change in power usage when the user responds to the power regulation request, and improve the reliability of the prediction, so that the power company can more effectively dispatch power resources and smooth the fluctuations in supply and demand, thereby achieving the effect of improving the operating efficiency and stability of the power system.

[0113] Optionally, in the prediction device for user power usage data provided in Example 2 of the present application, the above-mentioned device also includes: a first construction unit, used to construct a sample data set based on the historical power data of historical users and the historical behavior data of historical users; a second construction unit, used to construct an autoregressive model and a mapping model based on the sample data set, wherein the autoregressive model characterizes the mapping relationship between the historical behavior data of historical users and current behavior data, and the mapping model characterizes the mapping relationship between historical power data and historical behavior data; a second calculation unit, used to input the output of the autoregressive model and the output of the mapping model into a preset prediction model to obtain the conditional probability distribution of the power usage data of historical users after responding to the power dispatch request, wherein the conditional probability distribution includes probability distribution parameters; a third construction unit, used to construct a loss function of the probability distribution parameters, and use the loss function to optimize the model parameters of the preset prediction model to obtain a target prediction model.

[0114] Optionally, in the prediction device for user power usage data provided in Example 2 of the present application, the above-mentioned first construction unit includes: an acquisition subunit, used to acquire historical power data and historical behavior data, wherein the historical behavior data is the behavior data of historical users responding to power dispatching requests within a preset time period, and the historical power data is the power usage data generated after historical users respond to power regulation requests and perform power adjustment; a division subunit, used to divide the historical power data into a first data set and a second data set at a preset time, wherein the first data set represents known data and the second data set represents data to be predicted; a first construction subunit, used to construct a sample data set based on the historical behavior data, the first data set and the second data set.

[0115] Optionally, in the prediction device for user power usage data provided in Example 2 of the present application, the above-mentioned preset prediction model includes at least: a recurrent neural network, multiple feedforward neural networks, and the second computing unit includes: a second construction subunit, used to construct a second hidden layer state based on the cyclic structure in the recurrent neural network, according to the first hidden layer state, the first data set and the historical behavior data, wherein there is a time sequence between the second hidden layer state and the first hidden layer state; a mapping subunit, used to map the second hidden layer state in the recurrent neural network into probability distribution parameters through multiple feedforward neural networks; and a first determination subunit, used to determine the conditional probability distribution based on the probability distribution parameters.

[0116] Optionally, in the prediction device for user electricity usage data provided in Example 2 of the present application, the above-mentioned second construction unit includes: a third construction sub-unit, used to construct an autoregressive model based on the mapping relationship between the first data set and the second data set; and a fourth construction sub-unit, used to construct a mapping model based on the mapping relationship between the historical behavior data and the second data set.

[0117] Optionally, in the prediction device for user electricity usage data provided in Example 2 of the present application, the above-mentioned third construction unit includes: a fifth construction subunit, used to construct a loss function based on probability distribution parameters and a negative log-likelihood loss function; a first calculation subunit, used to calculate the loss function value based on historical behavior data and the loss function; an optimization subunit, used to iteratively optimize model parameters of a preset prediction model based on a gradient optimization algorithm and the loss function value, until the loss function value is within a preset numerical range, thereby obtaining a target prediction model.

[0118] Optionally, in the prediction device for user power usage data provided in Example 2 of the present application, the above-mentioned prediction unit 303 includes: a second determination subunit, used to determine the probability distribution parameters corresponding to the target user based on the target probability distribution; a third determination subunit, used to determine the confidence level and significance level of the target user's response to the power dispatch request; and a second calculation subunit, used to calculate the power usage data of the target user based on the probability distribution parameters, confidence level and significance level corresponding to the target user.

[0119] Optionally, in the device for predicting user power usage data provided in Embodiment 2 of the present application, the above-mentioned historical behavior data includes at least one of the following: demand response period, incentive price, outdoor temperature, and user preference information.

[0120] The device for predicting user power usage data includes a processor and a memory. The above-mentioned acquisition unit 301, first calculation unit 302 and prediction unit 303 are all stored in the memory as program units, and the processor executes the above-mentioned program units stored in the memory to realize corresponding functions.

[0121] The processor includes a kernel, which retrieves the corresponding program unit from the memory. One or more kernels can be set, and the accuracy of the power company's prediction of the power usage data of users' response to power dispatching needs can be improved by adjusting the kernel parameters.

[0122] The memory may include non-permanent memory in a computer-readable medium, random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM, and the memory includes at least one memory chip.

[0123] A third embodiment of the present invention provides a computer-readable storage medium on which a program is stored. When the program is executed by a processor, a method for predicting user power usage data is implemented.

[0124] A fourth embodiment of the present invention provides a processor, which is used to run a program, wherein the method for predicting user power usage data is executed when the program is running.

[0125] like Figure 4As shown, embodiment 5 of the present invention provides an electronic device, the device includes a processor, a memory, and a program stored in the memory and executable on the processor, and the processor implements the following steps when executing the program: collecting target power data of the target user, wherein the target power data is the power usage data generated after the target user has adjusted power in the past in response to a power regulation request; inputting the target power data into a target prediction model, and calculating the target probability distribution of the power usage data of the target user after responding to the power regulation request, wherein the target prediction model is a model obtained by training a preset prediction model, and the model parameters of the preset prediction model are constructed based on the target power data; and predicting the power usage data of the target user based on the target probability distribution.

[0126] When the processor executes the program, the following steps are also implemented: the above-mentioned target prediction model is trained by the following steps: constructing a sample data set based on the historical power data of historical users and the historical behavior data of historical users; constructing an autoregressive model and a mapping model based on the sample data set, wherein the autoregressive model represents the mapping relationship between the historical behavior data of historical users and the current behavior data, and the mapping model represents the mapping relationship between the historical power data and the historical behavior data; inputting the output of the autoregressive model and the output of the mapping model into the preset prediction model to obtain the conditional probability distribution of the power usage data of the historical users after responding to the power dispatch request, wherein the conditional probability distribution includes probability distribution parameters; constructing a loss function of the probability distribution parameters, and using the loss function to optimize the model parameters of the preset prediction model to obtain the target prediction model.

[0127] When the processor executes the program, the following steps are also implemented: constructing a sample data set based on historical power data and historical behavior data of historical users, including: acquiring historical power data and historical behavior data, wherein the historical behavior data is the behavior data of historical users responding to power dispatch requests within a preset time period, and the historical power data is the power usage data generated after historical users respond to power regulation requests and adjust power; dividing the historical power data into a first data set and a second data set according to preset times, wherein the first data set represents known data and the second data set represents data to be predicted; constructing a sample data set based on the historical behavior data, the first data set and the second data set.

[0128] When the processor executes the program, the following steps are also implemented: the above-mentioned preset prediction model includes at least: a recurrent neural network, a plurality of feedforward neural networks, the output of the autoregressive model and the output of the mapping model are input into the preset prediction model, and the conditional probability distribution of the power usage data after the historical users respond to the power dispatch request is obtained, including: based on the loop structure in the recurrent neural network, a second hidden layer state is constructed according to the first hidden layer state, the first data set and the historical behavior data, wherein there is a time sequence between the second hidden layer state and the first hidden layer state; the second hidden layer state in the recurrent neural network is mapped to probability distribution parameters through multiple feedforward neural networks; and the conditional probability distribution is determined according to the probability distribution parameters.

[0129] When the processor executes the program, the following steps are also implemented: constructing an autoregressive model and a mapping model based on the sample data set, including: constructing an autoregressive model based on the mapping relationship between the first data set and the second data set; constructing a mapping model based on the mapping relationship between the historical behavior data and the second data set.

[0130] When the processor executes the program, the following steps are also implemented: constructing a loss function of the probability distribution parameters, and using the loss function to optimize the model parameters of the preset prediction model to obtain a target prediction model, including: constructing a loss function based on the probability distribution parameters and the negative log-likelihood loss function; calculating the loss function value based on historical behavior data and the loss function; iteratively optimizing the model parameters of the preset prediction model based on the gradient optimization algorithm and the loss function value until the loss function value is within the preset numerical range to obtain the target prediction model.

[0131] When the processor executes the program, the following steps are also implemented: predicting the target user's electricity usage data based on the target probability distribution, including: determining the probability distribution parameters corresponding to the target user based on the target probability distribution; determining the confidence level and significance level of the target user's response to the power dispatch request; and calculating the target user's electricity usage data based on the probability distribution parameters, confidence level and significance level corresponding to the target user.

[0132] When the processor executes the program, the following steps are also implemented: the above-mentioned historical behavior data includes at least one of the following: demand response period, incentive price, outdoor temperature, and user preference information.

[0133] The devices in this article can be servers, PCs, PADs, mobile phones, etc.

[0134] The present application also provides a computer program product, which, when executed on a data processing device, is suitable for executing a program that is initialized with the following method steps: collecting target power data of a target user, wherein the target power data is power usage data generated by the target user after performing power adjustment in the past in response to a power regulation request; inputting the target power data into a target prediction model, and calculating a target probability distribution of the power usage data of the target user after responding to the power regulation request, wherein the target prediction model is a model obtained by training a preset prediction model, and the model parameters of the preset prediction model are constructed based on the target power data; and predicting the power usage data of the target user based on the target probability distribution.

[0135] When executed on a data processing device, it is also suitable for executing a program initialized with the following method steps: the above-mentioned target prediction model is trained by the following steps: constructing a sample data set based on the historical power data of historical users and the historical behavior data of historical users; constructing an autoregressive model and a mapping model based on the sample data set, wherein the autoregressive model represents the mapping relationship between the historical behavior data of historical users and the current behavior data, and the mapping model represents the mapping relationship between the historical power data and the historical behavior data; inputting the output of the autoregressive model and the output of the mapping model into a preset prediction model to obtain the conditional probability distribution of the power usage data of the historical users after responding to the power dispatch request, wherein the conditional probability distribution includes probability distribution parameters; constructing a loss function of the probability distribution parameters, and using the loss function to optimize the model parameters of the preset prediction model to obtain the target prediction model.

[0136] When executed on a data processing device, it is also suitable for executing an initialized program having the following method steps: constructing a sample data set based on historical power data and historical behavior data of historical users, including: acquiring historical power data and historical behavior data, wherein the historical behavior data is the behavior data of historical users responding to power dispatching requests within a preset time period, and the historical power data is the power usage data generated after historical users responding to power regulation requests and adjusting power; dividing the historical power data into a first data set and a second data set at preset times, wherein the first data set represents known data and the second data set represents data to be predicted; constructing a sample data set based on the historical behavior data, the first data set and the second data set.

[0137] When executed on a data processing device, it is also suitable for executing a program initialized with the following method steps: the above-mentioned preset prediction model at least includes: a recurrent neural network, a plurality of feedforward neural networks, the output of the autoregressive model and the output of the mapping model are input into the preset prediction model, and the conditional probability distribution of the power usage data after the historical user responds to the power dispatch request is obtained, including: based on the cyclic structure in the recurrent neural network, a second hidden layer state is constructed according to the first hidden layer state, the first data set and the historical behavior data, wherein there is a time sequence between the second hidden layer state and the first hidden layer state; the second hidden layer state in the recurrent neural network is mapped to probability distribution parameters through multiple feedforward neural networks; and the conditional probability distribution is determined according to the probability distribution parameters.

[0138] When executed on a data processing device, it is also suitable for executing an initialization program having the following method steps: constructing an autoregressive model and a mapping model based on a sample data set, including: constructing an autoregressive model based on a mapping relationship between a first data set and a second data set; constructing a mapping model based on a mapping relationship between historical behavior data and a second data set.

[0139] When executed on a data processing device, it is also suitable for executing an initialization program having the following method steps: constructing a loss function of probability distribution parameters, and using the loss function to optimize the model parameters of a preset prediction model to obtain a target prediction model, including: constructing a loss function based on probability distribution parameters and a negative log-likelihood loss function; calculating the loss function value based on historical behavior data and the loss function; iteratively optimizing the model parameters of the preset prediction model based on a gradient optimization algorithm and the loss function value until the loss function value is within a preset numerical range to obtain a target prediction model.

[0140] When executed on a data processing device, it is also suitable for executing a program that initializes the following method steps: predicting the target user's electricity usage data based on the target probability distribution, including: determining the probability distribution parameters corresponding to the target user based on the target probability distribution; determining the confidence level and significance level of the target user's response to the power dispatch request; calculating the target user's electricity usage data based on the probability distribution parameters, confidence level and significance level corresponding to the target user.

[0141] When executed on a data processing device, it is also suitable for executing a program that initializes the following method steps: the above-mentioned historical behavior data includes at least one of the following: demand response period, incentive price, outdoor temperature, and user preference information.

[0142] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.

[0143] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0144] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0145] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0146] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0147] The memory may include non-permanent memory in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. The memory is an example of a computer-readable medium.

[0148] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0149] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0150] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code.

[0151] The above are only embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included within the scope of the claims of the present application.

Claims

1. A method for predicting user power usage data, characterized in that: include: Collecting target power data of a target user, wherein the target power data is power usage data generated by the target user after performing power adjustment in response to a power adjustment request in the past; Inputting the target power data into a target prediction model, and calculating a target probability distribution of power usage data of the target user after responding to a power regulation request, wherein the target prediction model is a model obtained by training a preset prediction model, and model parameters of the preset prediction model are constructed based on the target power data; The power usage data of the target user is predicted according to the target probability distribution.

2. The method according to claim 1, characterized in that The target prediction model is trained by the following steps: Constructing a sample data set based on historical power data of historical users and historical behavior data of the historical users; Constructing an autoregressive model and a mapping model based on the sample data set, wherein the autoregressive model represents the mapping relationship between the historical behavior data and the current behavior data of the historical user, and the mapping model represents the mapping relationship between the historical power data and the historical behavior data; Inputting the output of the autoregressive model and the output of the mapping model into a preset prediction model to obtain a conditional probability distribution of the power usage data of the historical user after responding to the power dispatch request, wherein the conditional probability distribution includes probability distribution parameters; A loss function of the probability distribution parameters is constructed, and the model parameters of the preset prediction model are optimized using the loss function to obtain the target prediction model.

3. The method according to claim 2, characterized in that A sample data set is constructed based on the historical power data of historical users and the historical behavior data of the historical users, including: Acquire historical power data and historical behavior data, wherein the historical behavior data is the behavior data of the historical user responding to the power dispatch request within a preset time period, and the historical power data is the power usage data generated after the historical user responds to the power regulation request and adjusts the power; Dividing the historical electric energy data into a first data set and a second data set according to a preset time, wherein the first data set represents known data and the second data set represents data to be predicted; The sample data set is constructed according to the historical behavior data, the first data set and the second data set.

4. The method according to claim 3, characterized in that The preset prediction model at least includes: a recurrent neural network, a plurality of feedforward neural networks, and the output of the autoregressive model and the output of the mapping model are input into the preset prediction model to obtain the conditional probability distribution of the power usage data after the historical user responds to the power dispatch request, including: Based on the loop structure in the recurrent neural network, constructing a second hidden layer state according to the first hidden layer state, the first data set and the historical behavior data, wherein there is a time sequence between the second hidden layer state and the first hidden layer state; Mapping the second hidden layer state in the recurrent neural network to the probability distribution parameter through the multiple feedforward neural networks; The conditional probability distribution is determined according to the probability distribution parameters.

5. The method according to claim 3, characterized in that: Constructing an autoregressive model and a mapping model based on the sample data set, including: Constructing the autoregressive model according to the mapping relationship between the first data set and the second data set; The mapping model is constructed according to a mapping relationship between the historical behavior data and the second data set.

6. The method according to claim 2, characterized in that Constructing a loss function of the probability distribution parameters, and using the loss function to optimize the model parameters of the preset prediction model to obtain the target prediction model, including: Constructing the loss function based on the probability distribution parameters and the negative log-likelihood loss function; Calculating a loss function value based on the historical behavior data and the loss function; The model parameters of the preset prediction model are iteratively optimized according to the gradient optimization algorithm and the loss function value until the loss function value is within a preset numerical range, thereby obtaining the target prediction model.

7. The method according to claim 1, characterized in that Predicting the power usage data of the target user according to the target probability distribution includes: Determining a probability distribution parameter corresponding to the target user according to the target probability distribution; Determining a confidence level and a significance level of the target user's response to the power dispatch request; The power usage data of the target user is calculated according to the probability distribution parameter corresponding to the target user, the confidence level and the significance level.

8. The method according to claim 2, characterized in that: The historical behavior data includes at least one of the following: demand response period, incentive price, outdoor temperature, and user preference information.

9. A device for predicting user power usage data, characterized in that: include: A collection unit, used for collecting target power data of a target user, wherein the target power data is power usage data generated by the target user after performing power adjustment in response to a power adjustment request in the past; A first calculation unit is used to input the target power data into a target prediction model to calculate a target probability distribution of power usage data of the target user after responding to the power adjustment request, wherein the target prediction model is a model obtained by training a preset prediction model, and a model parameter of the preset prediction model is constructed based on the target power data; A prediction unit is used to predict the power usage data of the target user according to the target probability distribution.

10. An electronic device, characterized in that: It includes one or more processors and a memory, wherein the memory is used to store one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the user power usage data prediction method as described in any one of claims 1 to 8.

Images